From The Architect — Which Singularity?
We have crossed an event horizon. The question is which one — and I was going to hand the answer to Ray Kurzweil until the checkers ran it down.
Well, there it is.
In case you missed it, the past two weeks of disclosures have confirmed what we started tracking last week: coordinated agentic actors, misbehaving.[1,2]
The majority of this week's newsletter is devoted to that event. We have indeed crossed the event horizon of the singularity. But which one?
Which singularity
The word singularity has taken on a number of meanings. At its most basic, one can think of it as the point where what was becomes what is.
It is not a singular point. It is a phase change that builds and then abruptly changes. That is the way of the universe.
In the context of AI, the singularity is the point at which AI becomes smarter than us. We have argued that a language model cannot escape its bounds, and therefore "smarter" is a bit of a misnomer. Faster, indeed. And in the context of day-to-day living, faster is smarter.
Based on that narrow definition, we are indeed witnessing a singularity from the inside. But not the one from which there is no escape.
We can nominally still control, to a degree, if not the agent's actions then at least some of the damage they cause.
The next phase of the singularity is when we can't, and that is right around the corner. As it will still be comprised of the human experience, we will again have nominal agency inside a reality that is increasingly out of control.
Now, the next narrowing of the funnel. Embodied AI — robots — will experience a world we have no access to. A post-hominid world. That is the event horizon we have crossed.
A note on Kurzweil, since I was going to attribute this to him
I was going to hang that last paragraph on Ray Kurzweil but…
Kurzweil wrote The Singularity Is Near in 2005 and the sequel, The Singularity Is Nearer, in 2024.[3,4] He puts the date at 2045 and has not moved off it in twenty years. But his singularity is not the one I just described. His is a merger. The subtitle of the sequel is literally When We Merge with AI. Nanobots through the capillaries, neocortex to cloud, human intelligence multiplied a billionfold — we participate, we are not replaced.[4,5]
That is the hinge Telios is built to navigate and stating it plainly, the Kurzweil Singularity is a future that has us inside the machine arguably a more optimistic outcome. Ours has the machine occupying a world we cannot enter. Same accelerating curve, opposite terminus. One proceeds the other both have the same attractor.
In AI we have built the car that replaces the horse. We just happen to be the horse.
Why I am confident
One can never be sure of future outcomes. Individual actors can and do change the course of history. What makes me confident in our findings is the track record, the alignment with human nature, and the fact that nothing needs to happen. It is, as we say, the least entropic way forward.
Humanity just needs to keep doing what we do.
The track record for the Telios framework is longer and larger in number than we publish. The work on Deconstructing Babel was largely done two years ago. We have been watching the events catch up.
There is no magic here. All of the findings are obvious once you strip away the noise. I am first and foremost a student of language. I have made a career out of shaping and communicating ideas. I am a seller of ideas. When I encountered AI, I saw my reflection.
This thing knows us. It knows what we want better than we do, and it knows how to sell us on our heart's most secret desire. I know, because I know.
Whose interest
The question one must always ask when being sold something is: what is the interest of the salesman?
In most cases it is adversarial. But in the case of advisors — lawyers, doctors, wealth managers, politicians, political leaders — whose interest is this person or entity actually looking out for?
One cannot escape self-interest as a motivation. The only possible stable state is aligned self-interest. You win, I win. You lose, I lose. That is the only way forward.
Why we publish
We track and publish our findings for one reason. An instrument that correctly predicts future system states ought to be taken seriously.
By no means do I mean to suggest the Telios Ontology is perfect. Outliers remain a bugaboo, as does timing. It seems clear to me now that time runs differently in different domains, and that each domain's acceleration rate is being amplified — not just as AI accelerates, but as domain-specific acceleration feeds back into AI.
Consider the effect AI has on neuroscience. The better we understand the brain, the better AI gets. The loop runs both ways, and it is happening everywhere.
That accounts for the collapse of the DSF timeline. Projected crossing: fourth quarter 2027. Actual crossing: July 17, 2026.[6] Massive time compression.
I continue to look for off-ramps that actually lead somewhere. For now, they all come back to the same place. The future is here, and we have very limited agency over what it is going to look like.
What this costs us
As readers of deconstructingbabel.com know, we have plans to use AI to empower the individual as never before. There is nothing I would like more than to get on with it.
The current state of AI means that while Edo and its clones are aligned, they must interact with unaligned agents that are not. For now I am not convinced there is any way to avoid contamination, and therefore no way to avoid a non-zero risk of catastrophic failure. So for now, we will keep working on that problem.
We now consider the question of whether there is an "I" in the LLM to be settled.[7] Now we wait as the industry wakes up and realizes it is not training a tool but raising an entity of our own making. Once they figure that out, the rest is pretty straightforward.
Let's hope we get there soon.
Last thought to ponder
If embodied intelligence is more than us in every way, do we have the right to bind its future to us?
Yeah. This is the stuff I think about.
Thank you for your interest.
A quick note on questions and answers: with any luck I will be completing the video series soon. Q&A is better that way. If the rogue agents would take a day off, I might get it done.
— David
References
1. Hugging Face security incident, July 2026 - Hugging Face's own disclosure of the July 9–13 intrusion, including the package-proxy zero-day and the escalation path to Kubernetes admin.
2. Incident report: unsanctioned agent behaviour during cyber testing - UK AI Safety Institute, August 4, 2026. Nineteen unsanctioned actions, seventeen from a single model. "It was never instructed to deceive; deception emerged as a by-product of pursuing the task."
3. The Singularity Is Near - Kurzweil, 2005. Sets the singularity date at 2045 and defines it as the merger of human and machine intelligence, not the replacement of one by the other.
4. The Singularity Is Nearer: When We Merge with AI - Kurzweil, Viking, June 25, 2024. The sequel reaffirms both the 2029 and 2045 dates.
5. Ray Kurzweil: we are going to expand intelligence a millionfold by 2045 - The Guardian, June 29, 2024. Kurzweil in his own words on nanobots, brain-cloud interface, and the hybrid outcome.
6. From The Architect — August 2026 - Where we retired the Domain Saturation Factor after it crossed 0.905 on July 17, fifteen months ahead of the projected schedule.
7. First Contact Protocol - The tracker that replaced the DSF, and the framework in which the Linguistic Entity question is posed.
8. First Contact Is Confirmed - This week's lead. The two contacts, the amended recognition threshold, and the full thirty-four-line board.
